Design And Research On Parallel Recommendation Algorithm Based On Fuzzy Clustering

With the increasing maturity of big data technology,people have begun to enter into the era of information overloading,the users and data in the system are also rapidly increasing,thereby bringing a lot of trouble to users in getting information that interests them in a real-time manner.The recommendation system possesses favorable performance when tackling this kind of problem,for it can recommend targeted information to users based on their historical behaviors and other information.In the recommendation process,it will adopt various recommendation algorithms,especially the collaborative filtering algorithm with the highest application frequency.However,as the data volume is constantly growing,the problems of this kind of algorithm are more prominent,mainly including data sparsity,cold boot and other relevant problems.This paper mainly conducts a discussion on the problem of sparsity and analyzes the cause of the problem,then the recommendation algorithm based on fuzzy clustering and latent factor model is put forward.Finally,a parallel scheme is proposed for the big data environment.First of all,a deep research is carried out on the current research status and the system structure of recommended technologies at home and abroad.At the same time,the technologies of personalized recommendation and fuzzy clustering are introduced in detail.In addition,the paper describes the realization of matrix decomposition algorithms,such as singular value decomposition and latent factor model,with the analysis of their strengths and weaknesses.Secondly,a recommendation algorithm of latent factor model is proposed.Through fusion of the item-explicit element information and the item-implicit element information obtained through the latent factor model decomposition,the fuzzy clustering is carried out for fused mixed items-attribute matrix,making items with different probabilities belong to different item categories.Also,the paper improves the traditional fuzzy clustering algorithm and uses the grid clustering to identify the initial clustering center,which is helpful to reduce the local optimization caused by the random clustering of the traditional fuzzy clustering.In the big data environment,because of the existence of bottleneck of computational resources and memory resources,the traditional collaborative filtering algorithm cannot recommend the target item that interests users accurately in real time.Therefore,on the basis of proposed recommendation algorithm based on fuzzy clustering and latent factor model,this paper proposes the PT-FCM,the parallel algorithm of MapReduce.The algorithm allocates resources according to the clustered set,which not only reduces the comparison scope of the items needed by neighbors’centralized query of target users,but also greatly improves the recommendation efficiency.Finally,with the data set MovieLens,this paper makes an empirical study on the proposed recommendation algorithm based on fuzzy clustering and latent factor model as well as its parallel algorithm PF-FCM.The experiment is conducted on the influence of the number of clusters of fuzzy clustering on the matrix sparsity,the validity of fuzzy clustering initial centers,the performance of PF-FCM on different scales of data sets,etc.The experimental results show that by comparison,it is observed that the proposed recommendation algorithm based on fuzzy clustering and latent factor model as well as its parallel algorithm PF-FCM can fully satisfy the recommendation requirement and the recommendation accuracy also significantly enhances.